Aerial Image Target Detection And Real-time Transmission Based On Improved YOLOv4 Algorit

With the development of machine vision,image target detection technology has been widely used in people's lives.At the same time,in order to meet the application requirements,the deep learning algorithm has made great progress,and put forward various models such as target detection,image fusion and image segmentation.In recent years,the emergence of unmanned aerial vehicle(UAV)provides a new perspective for machine vision,which can be flexibly applied to a variety of scenes because it is not limited by the equipment position.Therefore,the combination of machine vision and UAV technology will provide many conveniences in security monitoring,military reconnaissance,border monitoring and so on.However,UAV image acquisition,processing and transmission require high real-time performance and are limited by load.Therefore It is an urgent to reduce algorithm complexity,improve image processing speed and reduce system volume and quality.In this paper,the target detection algorithms at home and abroad are compared and analyzed.Considering that the target detection algorithm needs to meet the real-time requirements and can be implemented on embedded devices,Yolov4 algorithm,a one-stage model,is selected and improved,and a network layer simplification scheme for the target detection algorithm is proposed.Using a new special chip for image processing to build an image acquisition and compression transmission system.On the premise of ensuring the accuracy,the speed of image processing and transmission is improved to realize real-time detection.In view of the problems of small target object,high image resolution and high detection speed in aerial images,this paper proposes three improvement strategies based on YOLOv4.Firstly,the Canopy algorithm and K-Means++ clustering algorithm are used to generate Anchor Box,Which improves the stability and detection speed of the network.Secondly,Penalties are added to the confidence cross entropy loss function and category probability cross entropy loss function in YOLOv4 algorithm,which improves the detection accuracy of small targets;Finally,pruning the unimportant channel or network layer in the network ensure the recognition accuracy,reduces the computational complexity of the algorithm and improves the detection speed.In view of the problems of high speed of aerial image acquisition and high pressure of target detection and image transmission in aerial image target detection,this paper takes HI3559 A,an artificial intelligence chip of Hi Silicon,as the core,and then uses the Smart Platform and Media Process Platform provided by Hi Silicon to complete the circuit design of hardware module and the software design of image acquisition,target detection and compression coding,respective.It realizes high-speed processing and compression transmission of aerial images.Compared with traditional methodsm,it has the characteristics of small size,easy carrying and good real-time performance.